Method for detecting surface flatness of precast beam based on three-dimensional point cloud model

Abstract

The present invention discloses a method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model, including the following steps: (1) performing, according to a specific geometry of a three-dimensional point cloud model of a target component in a three-dimensional coordinate system, coarse calibration and fine calibration on the model sequentially to determine a spatial rotation matrix and perform point cloud coordinate calibration; (2) determining normal vectors at positions of points of the three-dimensional point cloud model of the component according to a principal component analysis method and a K-nearest-neighbor principle, so that a to-be-detected surface is segmented and extracted by defining a normal vector direction and a coordinate interval; and (3) iteratively searching for an optimal reference plane according to a form relationship between the to-be-detected surface and the three-dimensional coordinate system and calculating flatness of the surface.

Description

TECHNICAL FIELD

The present invention relates to the field of detecting manufacturing quality of precast beams in bridge engineering, and specifically, to a method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model.

BACKGROUND

As an important evaluation index for detecting the manufacturing quality of precast beams, flatness reflects the aesthetic effect of construction and has some impact on the mechanical properties of the precast beams during service. Flatness that fails to meet specifications represents unreliable manufacturing quality of the precast beams to some extent. Currently, in the methods for detecting the flatness of precast beam in practical projects, contact measurement methods such as feeler gauges and profilometers are usually used. Such methods have low detection efficiency and insufficient accuracy and produce detection results that are not easy to manage digitally. A three-dimensional point cloud model is a set of massive points that contain properties of a target surface. The spatial form of an object surface is reflected based on the three-dimensional coordinates of points. Surface feature information of a precast component is acquired by using a three-dimensional laser scanner. Compared with the conventional detection methods, the three-dimensional point cloud model has advantages such as a non-contact type, a high sampling rate, and a high degree of automation. However, based on the point cloud model that contains information data of massive three-dimensional coordinates, how to quickly extract a target surface from the entire model, how to accurately detect the flatness of the surface, and how to implement integration into automatic calculation steps need to be solved urgently.

SUMMARY

Objectives of the present invention: for a precast beam with a relatively regular geometric shape, a method for detecting surface flatness based on a three-dimensional point cloud model is provided, which is easy to be implemented by programming, and improves the calculation efficiency and the degree of automation compared with the conventional detection methods.

Technical solutions: a method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model is provided, and includes the following steps:

• • (1) performing, according to a specific geometry of a three-dimensional point cloud model of a target component in a three-dimensional coordinate system, coarse calibration and fine calibration on the model sequentially to determine a spatial rotation matrix and perform point cloud coordinate calibration;
  • (2) determining normal vectors at positions of points of the three-dimensional point cloud model of the component according to a principal component analysis method and a K-nearest-neighbor principle, so that a to-be-detected surface is segmented and extracted by defining a normal vector direction and a coordinate interval; and
  • (3) iteratively searching for an optimal reference plane according to a form relationship between the to-be-detected surface and the three-dimensional coordinate system and calculating flatness of the surface.

Preferably, step (1), step (2), and step (3) all need to satisfy constraint conditions: a geometric shape of a precast beam is regular, a placement position of the precast beam is horizontal, an origin of the three-dimensional coordinate system is set at a centroid of the point cloud model, and directions of an X-axis, a Y-axis, and a Z-axis of the three-dimensional coordinate system are respectively parallel to directions of a beam width, a beam length, and a beam height of the component.

Preferably, step (1) specifically includes the following steps:

• • 1.1 setting, for a three-dimensional point cloud model Pt₀ of the precast beam, an origin O of an original coordinate system X₀Y₀Z₀ at a centroid of the three-dimensional point cloud model:

${\mathrm{Pt}}_c=\left[\begin{array}{ccc}{x}_1-{\mu }_X& y_1-{\mu }_Y& z_1-{\mu }_Z\\ x_2-{\mu }_X& y_2-{\mu }_Y& z_2-{\mu }_Z\\ & ⋮& \\ x_n-{\mu }_X& y_n-{\mu }_Y& z_n-{\mu }_Z\end{array}\right],$

• • where Pt_c is a centralized three-dimensional point cloud model; x₁, . . . x_n are coordinate values of points X₀ in Pt₀, y₁ . . . y_n are coordinate values of Y₀, and z₁ . . . z_n are coordinate values of Z₀; and μ_X is an average value of the coordinate values of the points X₀ in Pt₀, μ_Y is an average value of the coordinate values of original Y₀, and μ_Z is an average value of the coordinate values of original Z₀;
  • 1.2 determining, for the precast beam placed horizontally, that a positive direction of an initial Z₁ coordinate axis is an opposite direction of gravity and is parallel to a direction of the beam height, making a projection of the three-dimensional point cloud model Pt_c onto a plane X₀OY₀, performing a principal component analysis on the projection to first decentralize a projected point cloud according to a principle of the principal component analysis, calculate a covariance matrix of the decentralized point cloud, and perform singular value decomposition on the covariance matrix to obtain a group of eigenvalues and an eigenvector uniquely corresponding to each eigenvalue, where an eigenvector corresponding to the largest eigenvalue is a first principal component, and an eigenvector corresponding to the second largest eigenvalue is a second principal component; and defining a direction of a Y₁ coordinate axis as a direction of the first principal component, and defining a direction of an X₁ coordinate axis as a direction of the second principal component, to complete calibration of an initial coordinate system X₁Y₁Z₁, that is, coarse calibration of coordinates of the three-dimensional point cloud model; and
  • 1.3 respectively making slices at appropriate positions capable of reflecting features of the beam width, the beam length, and the beam height of the component, where thicknesses of the slices are twice a point cloud density; making a projection of points contained in the slices to a slice plane and performing the principal component analysis, fitting the projection to a straight line, and comparing angles between the projected straight line and two coordinate axes; and defining a direction of a coordinate axis with a smaller angle with the projected straight line as the direction of the first principal component according to the principle of the principal component analysis, and defining a direction of an other coordinate axis in the slice plane as the direction of the second principal component, to complete calibration of a final coordinate system XYZ, that is, fine calibration of the coordinates of the three-dimensional point cloud model after the steps are completed for all the directions of the beam width, the beam length, and the beam height; and
  • calculating, according to an angle between the final coordinate system XYZ and the initial coordinate system X₁Y₁Z₁, rotation matrices R_x, R_y, and R_z corresponding to the point cloud that changes from the initial coordinate system X₁Y₁Z₁ to the final coordinate system XYZ, where if a rotation direction of rotating from the initial coordinate system X₁Y₁Z₁ to the final coordinate system XYZ is counterclockwise rotation, the obtained rotation matrices R_x, R_y and R_z are respectively:

$R_x=\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos \alpha & -\sin \alpha \\ 0& \sin \alpha & \cos \alpha \end{array}\right],$ $R_y=\left[\begin{array}{ccc}\cos \beta & 0& \sin \beta \\ 0& 1& 0\\ -\sin \beta & 0& \cos \beta \end{array}\right],\mathrm{and}$ $R_z=\left[\begin{array}{ccc}\cos \gamma & -\sin \gamma & 0\\ \sin \gamma & \cos \gamma & 0\\ 0& 0& 1\end{array}\right],$

• • and
  • if the rotation direction is clockwise rotation, the obtained rotation matrices R_x, R_y, and R_z are respectively:

$R_x=\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos \alpha & \sin \alpha \\ 0& -\sin \alpha & \cos \alpha \end{array}\right],$ $R_y=\left[\begin{array}{ccc}\cos \beta & 0& -\sin \beta \\ 0& 1& 0\\ \sin \beta & 0& \cos \beta \end{array}\right],\mathrm{and}$ $R_z=\left[\begin{array}{ccc}\cos \gamma & \sin \gamma & 0\\ -\sin \gamma & \cos \gamma & 0\\ 0& 0& 1\end{array}\right],$

• • where α is an angle between a coordinate system Y₁OZ₁ and a coordinate system YOZ; β is an angle between a coordinate system X₁OZ₁ and a coordinate system XOZ; and γ is an angle between a coordinate system X₁OY₁ and a coordinate system XOY; and obtaining a spatial rotation matrix R=R_x·R_y·R_z according to R_x, R_y, and R_z to implement point cloud coordinate calibration:Pt=R·Pt_c.
  • where Pt_c is a centralized three-dimensional point cloud model, and Pt is a three-dimensional point cloud model after coordinate calibration.

Preferably, step (2) specifically includes the following steps:

• • 2.1 selecting, for an i^th point p_i(x_i, y_i, z_i) in the three-dimensional point cloud model Pt, K nearest neighbor points according to the K-nearest-neighbor principle:P_i^K={p_i¹,p_i² . . . p_i^K|i∈{1,2 . . . N},K∈Z},
  • where N is the number of points in the three-dimensional point cloud model Pt, P_i^K is a set of K nearest neighbor points of a point p_i, and p_i^K is a k^th neighbor point; and
  • sequentially performing decentralization, covariance matrix calculation, and singular value decomposition on P_i^K according to the principle of the principal component analysis, defining a normal vector {right arrow over (n₁)} at a point p_i(x_i, y_i, z_i) as an eigenvector corresponding to a minimum eigenvalue, and specifying that a direction of the normal vector {right arrow over (n₁)} indicates a positive direction of a coordinate axis;
  • 2.2 rotating a to-be-detected surface A that requires surface flatness detection in the precast beam to be parallel to a coordinate plane B:A∥B,
  • where the coordinate plane B is any plane of XOY, XOZ, and YOZ; setting a reference vector {right arrow over (n_r)} as a unit vector perpendicular to the coordinate plane B, and calculating angles between normal vectors at points in the three-dimensional point cloud model Pt and the reference vector based on a vector angle formula, where an angle θ_i between a normal vector {right arrow over (n_i)} at the i^th point p_i(x_i, y_i, z_i) in the three-dimensional point cloud model Pt and the reference vector {right arrow over (n_r)} is:

${\theta }_i=\arccos \left(\frac{\overrightarrow{n_i}·\overrightarrow{n_r}}{❘\overrightarrow{n_i}❘❘\overrightarrow{n_r}❘}\right);$

• • setting an angle threshold θ_s as |θ_s|≤5°, where if a constraint condition θ_i≤θ_s is satisfied during calculation, that is, θ_i is not greater than the angle threshold θ_s, the point p_i(x_i, y_i, z_i) is a point forming the to-be-detected surface A, and the point is extracted into a set A₁; making, for points in the set A₁, a histogram by taking coordinate values of the points on a coordinate axis perpendicular to the to-be-detected surface A; and extracting points contained in an interval with a maximum frequency number into a set A₂; and
  • 2.3 selecting, for an h^th point p_h(x_h, y_h, z_h) in the set A₂, K nearest neighbor points around the point to form a set P_h^K according to the K-nearest-neighbor principle; defining a point distance d_h of the point p_h as an average value of distances between the point and points in the set P_h^K, and calculating an average value d and a standard deviation σ of point distances in the set A₂ if a result is a Gaussian distribution:

$\stackrel{\_}{d}=\frac{1}{N_{A_2}}\sum _{h=1}^{N_{A_2}}{d}_h,\sigma =\sqrt{\frac{1}{N_{A_2}}\sum _{h=1}^{N_{A_2}}{\left(d_h-\stackrel{\_}{d}\right)}^2},h\in \left\{1,2\dots N_{A_2}\right\},$

• • where N_A2 is the number of points in the set A₂; and setting d+2σ as a distance threshold, where if a constraint condition: d_h≤d+2σ is satisfied during calculation, that is, the point distance is not greater than the distance threshold, the point p_h(x_h, y_h, z_h) is a point forming the to-be-detected surface, and is extracted into the to-be-detected surface A.

Preferably, step (3) specifically includes the following steps:

• • 3.1 because the to-be-detected surface A is formed by three-dimensional coordinate points and has a particular thickness, making, by passing through a point nearest to the coordinate plane B in the to-be-detected surface A, a reference plane RP₁ parallel to the coordinate plane B, the reference plane RP₁ being a₁x+b₁y+c₁z+d₁=0; and making, by passing through a point farthest from the coordinate plane B in the to-be-detected surface A, a reference plane RP_n parallel to the coordinate plane B, the reference plane RP_n being a_nx+b_ny+c_nz+d_n=0, where a₁, b₁, c₁, d₁, a_n, b_n, c_n, and d_n are coefficients;
  • setting an iteration distance t as a point cloud density in the to-be-detected surface A, gradually translating the reference plane from RP₁ to RP_n by an increment of t, and calculating distances between points in the to-be-detected surface A and reference planes at each moving moment, where a Euclidean distance r_j between a j^th point p_j(x_j, y_j, z_j) in the to-be-detected surface A and an m^th reference plane RP_m is:

$r_j=❘\frac{a_m{x}_j+b_m{y}_j+c_m{z}_j+d_m}{\sqrt{a_m^2+b_m^2+c_m^2}}❘,j\in \left\{1,2\dots N_A\right\},m\in \left\{1,2\dots N_n\right\},$

• • where N_A is the number of points in the to-be-detected surface A, N_n is the number of reference planes, an expression of the reference plane RP_m is a_mx+b_my+c_mz+d_m=0, and a_m, b_m, c_m, and d_m are coefficients; and defining, according to a least absolute deviation method, an optimal reference plane RP_A as a corresponding reference plane when a sum of r_j of the points in the to-be-detected surface A is minimal:

${\mathrm{RP}}_A\left(r_j\right)=\mathrm{argmin}\sum _{j=1}^{N_A}{r}_j,$

• • where an expression of the optimal reference plane RP_A is a_Ax+b_Ay+c_Az+d_A=0, and a_A, b_A, c_A, and d_A are coefficients; and
  • 3.2 defining, for the j^th point p_j(x_j, y_j, z_j) in the to-be-detected surface A, surface flatness f_j at the point as a Euclidean distance between the point p_j and the optimal reference plane RP_A:

$f_j=❘\frac{a_A{x}_j+b_A{y}_j+c_A{z}_j+d_A}{\sqrt{a_A^2+b_A^2+c_A^2}}❘,j\in \left\{1,2\dots N_A\right\};$

• • obtaining a flatness distribution map of the to-be-detected surface A through marking with different colors according to a value of f_j, to complete detection of surface flatness of the precast beam.

Beneficial effects: Compared with the existing technology, the technical solution of the present invention has the following beneficial technical effects:

The present invention implements quick and accurate detection of flatness of a precast beam by performing coordinate calibration, target surface segmentation, and determination of an optimal reference plane on a three-dimensional point cloud model of the precast beam, which is beneficial to improving the calculation efficiency and the degree of automation. Avoiding the problem that the conventional contact measurement is prone to errors and results of the conventional contact measurement are less integrated, thereby saving calculation time and improving the degree of digital integration of detection results.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a flowchart of a method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model according to the present invention;

FIG. 2 is a schematic diagram of implementing fine calibration of coordinate axes according to slice features;

FIG. 3 is a schematic diagram of a normal vector n_i at an i^th point p_i;

FIG. 4 is a schematic diagram of a histogram of a surface feature; and

FIG. 5 is a schematic diagram of a process of searching for an optimal reference plane.

DETAILED DESCRIPTION

The technical solution of the present invention is further described below with reference to the drawings and specific embodiments.

The present invention needs to satisfy the following constraint conditions:

• • (1) A precast beam has a regular geometric shape. Generally, shapes of precast beams are regular basic patterns, mainly a rectangle, an I-shaped figure, etc. The present invention describes the specific implementation method by taking a cuboid beam with a square cross-section as an example.
  • (2) A placement position of the precast beam is basically horizontal, that is, there is a relatively small deviation between a direction of a Z-coordinate axis of a three-dimensional point cloud model and a direction of a beam height, which facilitates implementation of coordinate calibration.
  • (3) An origin of a three-dimensional coordinate system is set at a centroid of the three-dimensional point cloud model, a direction of an X-coordinate axis is parallel to a direction of a beam width, a direction of a Y-coordinate axis is parallel to a direction of a beam length, and the direction of the Z-coordinate axis is an opposite direction of gravity and is parallel to the direction of the beam height. For a regular precast beam model, the three-dimensional coordinate system is calibrated according to geometric features, making calculation and programming more convenient.

As shown in FIG. 1, a method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model in the present invention mainly includes the following steps:

• • (1) performing, according to a specific geometry of a three-dimensional point cloud model of a target component in a three-dimensional coordinate system, coarse calibration and fine calibration on the model sequentially to determine a spatial rotation matrix and perform point cloud coordinate calibration;
  • (2) determining normal vectors at positions of points of the three-dimensional point cloud model of the component according to a principal component analysis method and a K-nearest-neighbor principle, so that a to-be-detected surface is segmented and extracted by defining a normal vector direction and a coordinate interval; and
  • (3) iteratively searching for an optimal reference plane according to a form relationship between the to-be-detected surface and the three-dimensional coordinate system and calculating flatness of the surface.

In this embodiment, step 1 specifically includes the following steps.

• • 1.1 For a three-dimensional point cloud model Pt₀ of a precast beam, an origin O of an original coordinate system X₀Y₀Z₀ is set at a centroid of the three-dimensional point cloud model:

${\mathrm{Pt}}_c=\left[\begin{array}{ccc}{x}_1-{\mu }_X& y_1-{\mu }_Y& z_1-{\mu }_Z\\ x_2-{\mu }_X& y_2-{\mu }_Y& z_2-{\mu }_Z\\ & ⋮& \\ x_n-{\mu }_X& y_n-{\mu }_Y& z_n-{\mu }_Z\end{array}\right],$

• • where Pt_c is a centralized three-dimensional point cloud model; x₁ . . . x_n are coordinate values of points X₀ in Pt₀, y₁ . . . y_n are coordinate values of Y₀, and z₁ . . . z_n are coordinate values of Z₀; and μ_X is an average value of the coordinate values of the points X₀ in Pt₀, μ_Y is an average value of the coordinate values of original Y₀, and μ_Z is an average value of the coordinate values of original Z₀; the algorithm of the present invention mainly targets a precast beam with a regular geometric shape, and an origin of a coordinate axis is set at a position of a centroid of the model, which can effectively simplify calculation and improve the efficiency.
  • 1.2 The algorithm of the present invention mainly targets the precast beam placed horizontally. Therefore, it may be approximately determined that a positive direction of an initial Z₁ coordinate axis is the opposite direction of gravity and is approximately parallel to the direction of the beam height. A projection of the three-dimensional point cloud model Pt_c is made on a plane X₀OY₀, and a principal component analysis is performed on the projection. A projected point cloud is first decentralized according to a principle of the principal component analysis, a covariance matrix of the decentralized point cloud is calculated, and singular value decomposition is performed on the covariance matrix to obtain a group of eigenvalues and an eigenvector uniquely corresponding to each eigenvalue, where an eigenvector corresponding to the largest eigenvalue is a first principal component, and an eigenvector corresponding to the second largest eigenvalue is a second principal component. A direction of a Y₁ coordinate axis is defined as a direction of the first principal component, and a direction of an X₁ coordinate axis is defined as a direction of the second principal component, to complete calibration of an initial coordinate system X₁Y₁Z₁, that is, coarse calibration of coordinates of the three-dimensional point cloud model.
  • 1.3 Referring to FIG. 2, slices are respectively made at appropriate positions capable of reflecting features of the beam width, the beam length, and the beam height of the component, where thicknesses of the slices are twice a point cloud density, and the obtained slices may have the following characteristics: a. the slices have a sufficient length; and b. It should be ensured as much as possible that points contained in the slices is clear and continuous. Short slices or insufficient continuity may easily lead to identification errors during the principal component analysis.

A projection of the points contained in the slices is made to a slice plane, the principal component analysis is performed, and the projection of the points may be approximated as a straight line. By comparing angles between the projected straight line and two coordinate axes according to the principle of the principal component analysis, a direction of a coordinate axis which has a smaller angle with the projected straight line is defined as the direction of the first principal component, and a direction of an other coordinate axis in the slice plane is defined as the direction of the second principal component. After the steps for the directions of the beam width, the beam length, and the beam height are completed, calibration of a final coordinate system XYZ, that is, fine calibration of the coordinates of the three-dimensional point cloud model, is completed.

According to an angle between the final coordinate system XYZ and the initial coordinate system X₁Y₁Z₁, rotation matrices R_x, R_y, and R_z corresponding to the point cloud that changes from the initial coordinate system X₁Y₁Z₁ to the final coordinate system XYZ are calculated. If a rotation direction of rotating from the initial coordinate system X₁Y₁Z₁ to the final coordinate system XYZ is counterclockwise rotation, the obtained rotation matrices R_x, R_y, and R_z are respectively:

$R_x=\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos \alpha & -\sin \alpha \\ 0& \sin \alpha & \cos \alpha \end{array}\right],$ $R_y=\left[\begin{array}{ccc}\cos \beta & 0& \sin \beta \\ 0& 1& 0\\ -\sin \beta & 0& \cos \beta \end{array}\right],\mathrm{and}$ $R_z=\left[\begin{array}{ccc}\cos \gamma & -\sin \gamma & 0\\ \sin \gamma & \cos \gamma & 0\\ 0& 0& 1\end{array}\right],$ and

• • if the rotation direction is clockwise rotation, the obtained rotation matrices R_x, R_y, and R_z are respectively:

$R_x=\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos \alpha & \sin \alpha \\ 0& -\sin \alpha & \cos \alpha \end{array}\right],$ $R_y=\left[\begin{array}{ccc}\cos \beta & 0& -\sin \beta \\ 0& 1& 0\\ \sin \beta & 0& \cos \beta \end{array}\right],\mathrm{and}$ $R_z=\left[\begin{array}{ccc}\cos \gamma & \sin \gamma & 0\\ -\sin \gamma & \cos \gamma & 0\\ 0& 0& 1\end{array}\right],$

• • where α is an angle between a coordinate system Y₁OZ₁ and a coordinate system YOZ; β is an angle between a coordinate system X₁OZ₁ and a coordinate system XOZ; and γ is an angle between a coordinate system X₁OY₁ and a coordinate system XOY.

In summary, a spatial rotation matrix R=R_x·R_y·R_z is obtained, to implement point cloud coordinate calibration:Pt=R·Pt_c,

• • where Pt_c is a centralized three-dimensional point cloud model, and Pt is a three-dimensional point cloud model after coordinate calibration.

In this embodiment, step 2 specifically includes the following steps.

• • 2.1 Referring to FIG. 3, for an i^th point p_i(x_i, y_i, z_i) in a three-dimensional point cloud model Pt, K nearest neighbor points are selected according to the K-nearest-neighbor principle:P_i^K={p_i¹,p_i² . . . p_i^K|i∈{1,2 . . . N},K∈Z},
  • where N is the number of points in the three-dimensional point cloud model Pt, P_i^K is a set of K nearest neighbor points of a point p_i, and p_i^K is neighbor points.

Decentralization, covariance matrix calculation, and singular value decomposition are sequentially performed on P_i^K according to the principle of the principal component analysis, a normal vector {right arrow over (n₁)} at a point p_i(x_i, y_i, z_i) is defined as an eigenvector corresponding to a minimum eigenvalue, and it is specified that a direction of the normal vector {right arrow over (n₁)} indicates a positive direction of a coordinate axis. The principle is to fit a plane based on the K neighbor points of the point p_i(x_i, y_i, z_i), and convert calculation of the normal vector at the point p_i into calculation of a normal vector of the above fitted plane.

• • 2.2 A to-be-detected surface A that requires surface flatness detection in the precast beam is rotated to be parallel to a coordinate plane B:A∥B.
  • where the coordinate plane B is any plane of XOY, XOZ, and YOZ. A reference vector {right arrow over (n_r)} is set as a unit vector perpendicular to the coordinate plane B, and angles between normal vectors at points in the three-dimensional point cloud model Pt and the reference vector are calculated based on a vector angle formula, where an angle θ_i between a normal vector {right arrow over (n₁)} at the i^th point p_i(x_i, y_i, z_i) in the three-dimensional point cloud model Pt and the reference vector is:

${\theta }_i=\arccos \left(\frac{\overrightarrow{n_i}·\overrightarrow{n_r}}{❘\overrightarrow{n_i}❘❘\overrightarrow{n_r}❘}\right).$

An angle threshold θ_s is set as |θ_s|≤5°. If a constraint condition θ_i≤θ_s is satisfied during calculation, that is, θ_i is not greater than the angle threshold θ_s, the point p_i(x_i, y_i, z_i) is a point forming the to-be-detected surface A, and the point is extracted into a set A₁.

Referring to FIG. 4, for points in the set A₁, a histogram is made by taking coordinate values of the points on a coordinate axis perpendicular to the to-be-detected surface A, and points included in an interval with a maximum frequency number are extracted into a set A₂.

• • 2.3 For an h^th point p_h(x_h, y_h, z_h) in the set A₂, K nearest neighbor points around the point are selected to form a set P_h^K according to the K-nearest-neighbor principle. A point distance d_h of the point p_h is defined as an average value of distances between the point and points in the set P_h^K, and an average value d and a standard deviation σ of point distances in the set A₂ are calculated if a result is a Gaussian distribution:

$\stackrel{\_}{d}=\frac{1}{N_{A_2}}\sum _{h=1}^{N_{A_2}}{d}_h,\sigma =\sqrt{\frac{1}{N_{A_2}}\sum _{h=1}^{N_{A_2}}{\left(d_h-\stackrel{\_}{d}\right)}^2},h\in \left\{1,2\dots N_{A_2}\right\},$

• • where N_A2 is the number of points in the set A₂. d+2σ is set as a distance threshold, where if a constraint condition: d_h≤d+2σ is satisfied during calculation, that is, the point distance is not greater than the distance threshold, the point p_h(x_h, y_h, z_h) is a point forming the to-be-detected surface, and is extracted into the to-be-detected surface A.

In this embodiment, step 3 specifically includes the following steps.

• • 3.1 Because the to-be-detected surface A is formed by three-dimensional coordinate points and has a particular thickness, a reference plane RP₁(a₁x+b₁y+c₁z+d₁=0) parallel to the coordinate plane B is made by passing through a point nearest to the coordinate plane B in the to-be-detected surface A; and a reference plane RP_n(a_nx+b_ny+c_nz+d_n=0) parallel to the coordinate plane B is made by passing through a point farthest from the coordinate plane B in the to-be-detected surface A.

Referring to FIG. 5, an iteration distance t is set as a point cloud density in the to-be-detected surface A, the reference plane is gradually translated from RP₁ to RP_n by an increment of t, and distances between points in the to-be-detected surface A and reference planes are calculated at each moving moment, where a Euclidean distance r_j between a j^th point p_j(x_j, y_j, z_j) in the to-be-detected surface A and an m^th reference plane RP_m(a_mx+b_my+c_mz+d_m=0) is:

$r_j=❘\frac{a_m{x}_j+b_m{y}_j+c_m{z}_j+d_m}{\sqrt{a_m^2+b_m^2+c_m^2}}❘,j\in \left\{1,2\dots N_A\right\},m\in \left\{1,2\dots N_n\right\},$

• • where N_A is the number of points in the to-be-detected surface A, and N_n is the number of reference planes.

According to a least absolute deviation method, an optimal reference plane RP_A(a_Ax+b_Ay+c_Az+d_A=0) is defined as a corresponding reference plane when a sum of r_j of the points in the to-be-detected surface A is minimal:

${\mathrm{RP}}_A\left(r_j\right)=\mathrm{argmin}\sum _{j=1}^{N_A}{r}_j.$

• • 3.2 For the j^th point p_j(x_j, y_j, z_j) in the to-be-detected surface A, surface flatness f_j at the point is defined as a Euclidean distance between the point p_j and the optimal reference plane RP_A(a_Ax+b_Ay+c_Az+d_A=0):

$f_j=❘\frac{a_A{x}_j+b_A{y}_j+c_A{z}_j+d_A}{\sqrt{a_A^2+b_A^2+c_A^2}}❘,j\in \left\{1,2\dots N_A\right\}.$

A flatness distribution map of the to-be-detected surface A may be obtained through marking with different colors according to a value of f_j, to complete detection of surface flatness of the precast beam.

Because the algorithm is established based on a plurality of constraint conditions, the algorithm is not suitable for detection of surface flatness of all precast beams. However, for most of the commonly used types of precast beams in practical engineering, the algorithm can greatly improve the calculation efficiency and the automation degree of the flatness detection link compared with the existing conventional detection methods.

The following further describes the technical solutions of the present invention with reference to specific cases. The case calculation mainly performs algorithm programming based on MATLAB, to obtain calculation results.

The studied precast beam is a cuboid beam with a square cross-section, a beam length is 10 m, a beam width is 2 m, a beam height is 2 m, and a point cloud density is 1/cm². An area that has a width of 0.1 m and a length of 0.2 m and is 0.003 m above an original plane is preset at an end of a top surface of the beam, to simulate the situation that the precast beam is not flat. Rotation matrices are applied in advance:

$R_x=\left[\begin{array}{ccc}1& 0& 0\\ 0& 1& 0.0015\\ 0& -0.0015& 1\end{array}\right],$ $R_y=\left[\begin{array}{ccc}1& 0& -0.0042\\ 0& 1& 0\\ 0.0042& 0& 1\end{array}\right],\mathrm{and}$ $R_z=\left[\begin{array}{ccc}1& 0.0029& 0\\ -0.0029& 1& 0\\ 0& 0& 1\end{array}\right].$

• • 1. Decentralization is performed on a three-dimensional point cloud model of the precast beam, and coarse calibration is performed according to an initial form of the three-dimensional point cloud model, to obtain a rotation matrix

${\mathrm{Rot}}_{z1}=\left[\begin{array}{ccc}1& -0.0017& 0\\ 0.0017& 1& 0\\ 0& 0& 1\end{array}\right].$ Subsequently, according to step 1.3, fine calibration is respectively performed on directions of three coordinate axes in the three-dimensional point cloud model, to obtain rotation matrices:

${\mathrm{Rot}}_x=\left[\begin{array}{ccc}1& 0& 0\\ 0& 1& -0.0015\\ 0& 0.0015& 1\end{array}\right],$ ${\mathrm{Rot}}_y=\left[\begin{array}{ccc}1& 0& 0.0042\\ 0& 1& 0\\ -0.0042& 0& 1\end{array}\right],\mathrm{and}$ ${\mathrm{Rot}}_{z2}=\left[\begin{array}{ccc}1& -0.0012& 0\\ 0.0012& 1& 0\\ 0& 0& 1\end{array}\right].$

The rotation matrices are consistent with the preset rotation matrices.

• • 2. According to step 2.1, normal vectors at points are calculated. To extract a point cloud contained in the top surface, a reference normal vector {right arrow over (n_r)}=(0,0,1) is set, and angles between the normal vectors at the points and the reference normal vector are calculated. An angle threshold θ_s is set to be greater than or equal to 5°, it is sequentially determined whether the normal vectors at the points satisfy a standard, and points that satisfy a determining condition θ_i≤θ_s are extracted into a set A₂. A histogram is made with a width of 0.1 m by selecting z-coordinates of the points in the set A₂, and points corresponding to units that include more than 5000 three-dimensional points whose z-coordinates are greater than 0 are extracted. According to step 2.3, point distances, an average point distance, and a standard deviation of the extracted points are calculated, and points exceeding a distance threshold d+2σ are filtered out, to obtain a final point cloud of the top surface.
  • 3. According to step 3.1, a reference plane z=min z is made, and is moved to z=max z by an increment of 0.0001 m. For each reference plane z=z_i, distances between the points and the reference plane are calculated. An optimal reference plane is finally determined as z=0.9991, and a distance between a non-flat part and the optimal reference plane is 0.003 m, and is consistent with the initial setting.

The present invention relates to a list of symbols and parameters as follows:

	Centralized three-		Original		Initial
	dimensional point		coordinate		coordinate
Ptc	cloud model	X0Y0Z0	system	X1Y1Z1	system
xi	X-coordinate value of a	yi	Y-coordinate value	zi	Z-coordinate value
	point i		of the point i		of the point i
Rx	Rotation matrix around	Ry	Rotation matrix	Rz	Rotation matrix
	an X-axis		around a Y-axis		around a Z-axis
α	Angle between a	β	Angle between a	γ	Angle between a
	coordinate system		coordinate system		coordinate system
	Y1OZ1 and a		X1OZ1 and a		X1OY1 and a
	coordinate system		coordinate system		coordinate system
	YOZ		XOZ		XOY
K	The number of nearest	pi	ith three-
	neighbor points		dimensional
			coordinate point
PiK	Set of K nearest	piK	Neighbor point
	neighbor points of the
	ith pointp;
pij	jth neighbor point of pi	pc	Midpoint of K	A	To-be-detected
			neighbor points		surface
{right arrow over (nr)}	Reference vector	{right arrow over (n1)}	Normal vector at an	θi	angle between the
			ith point		normal vector {right arrow over (ni)}
					and the reference
					vector
θs	Angle threshold	A1	Set of points at	A2	Set of points
			which normal		whose coordinates
			vectors satisfy a		are in a specified
			threshold		range
di	Point distance	d	Average value of	σ	Standard deviation
			point distances		of the point
					distances
N	The number of points	NA	The number of	NA2	The number of
	in the three-		points in a set A		points in a set A2
	dimensional point
	cloud model
RPA	Optimal reference	zr	z-coordinate value	fi	Surface flatness at
	plane of the surface A		corresponding to a		the point i
			reference plane

The foregoing content further describes the present invention in detail with reference to only one specific embodiment, and it should not be considered that the present invention is limited to these descriptions. A person of ordinary skill in the art can further make simple deductions or replacements without departing from the concept of the present invention, and such deductions or replacements should all be considered as falling within the protection scope of the present invention.

Claims

1. A method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model, comprising the following steps: (1) performing, according to a specific geometry of a three-dimensional point cloud model of a target component in a three-dimensional coordinate system, coarse calibration and fine calibration on the model sequentially to determine a spatial rotation matrix and perform point cloud coordinate calibration;(2) determining normal vectors at positions of points of the three-dimensional point cloud model of the component according to a principal component analysis method and a K-nearest-neighbor principle, so that a to-be-detected surface is segmented and extracted by defining a normal vector direction and a coordinate interval; and(3) iteratively searching for an optimal reference plane according to a form relationship between the to-be-detected surface and the three-dimensional coordinate system and calculating flatness of the surface,wherein step (1) specifically comprises the following steps:1.1 setting, for a three-dimensional point cloud model Pt0 of the precast beam, an origin O of an original coordinate system X0Y0Z0 at a centroid of the three-dimensional point cloud model:

2. The method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model according to claim 1, wherein step (1), step (2), and step (3) all need to satisfy constraint conditions: a geometric shape of a precast beam is regular, a placement position of the precast beam is horizontal, an origin of the three-dimensional coordinate system is set at a centroid of the point cloud model, and directions of an X-axis, a Y-axis, and a Z-axis of the three-dimensional coordinate system are respectively parallel to directions of a beam width, a beam length, and a beam height of the component.

3. The method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model according to claim 2, wherein step (2) specifically comprises the following steps: 2.1 selecting, for an ith point pi(xi, yi, zi) in the three-dimensional point cloud model Pt, K nearest neighbor points according to the K-nearest-neighbor principle: PiK={pi1,pi2 . . . piK|i∈{1,2 . . . N},K∈Z}, wherein N is the number of points in the three-dimensional point cloud model Pt, PiK is a set of K nearest neighbor points of a point pi, and piK is a kth neighbor point; andsequentially performing decentralization, covariance matrix calculation, and singular value decomposition on PiK according to the principle of the principal component analysis, defining a normal vector ni at a point pi(xi, yi, zi) as an eigenvector corresponding to a minimum eigenvalue, and specifying that a direction of the normal vector {right arrow over (n)}i indicates a positive direction of a coordinate axis;2.2 rotating a to-be-detected surface A that requires surface flatness detection in the precast beam to be parallel to a coordinate plane B: A∥B, wherein the coordinate plane B is any plane of XOY, XOZ, and YOZ;setting a reference vector {right arrow over (n)}r as a unit vector perpendicular to the coordinate plane B, and calculating angles between normal vectors at points in the three-dimensional point cloud model Pt and the reference vector based on a vector angle formula, wherein an angle θi between a normal vector {right arrow over (n)}i at the ith point pi(xi, yi, zi) in the three-dimensional point cloud model Pt and the reference vector {right arrow over (n)}r is:

4. The method for detecting surface flatness of a precast beam based on a three-dimensional point cloud model according to claim 3, wherein step (3) specifically comprises the following steps: 3.1 because the to-be-detected surface A is formed by three-dimensional coordinate points and has a particular thickness, making, by passing through a point nearest to the coordinate plane B in the to-be-detected surface A, a reference plane RP1 parallel to the coordinate plane B, the reference plane RP1 being a1x+b1y+c1z+d1=0; and making, by passing through a point farthest from the coordinate plane B in the to-be-detected surface A, a reference plane RPn parallel to the coordinate plane B, the reference plane RPn being anx+bny+cnz+dn=0, wherein a1, b1, c1, d1, an, bn, cn, and do are coefficients;setting an iteration distance t as a point cloud density in the to-be-detected surface A, gradually translating the reference plane from RP1 to RPn by an increment of t, and calculating distances between points in the to-be-detected surface A and reference planes at each moving moment, wherein a Euclidean distance r between a jth point pj(xj, yj, zj) in the to-be-detected surface A and an mth reference plane RPm is: